High Speed Simulation and Freeform Optimization of Nanophotonic Devices with Physics-Augmented Deep Learning

Abstract

We introduce WaveY-Net, a hybrid data- and physics-augmented convolutional neural network that can predict electromagnetic field distributions with ultrafast speeds and high accuracy for entire classes of dielectric nanophotonic structures. This accuracy is achieved by training the neural network to learn only the magnetic near-field distributions of a system and to use a discrete formalism of Maxwell’s equations in two ways: to calculate electric fields from the magnetic fields and as physical constraints in the loss function. We show that WaveY-Net can accurately predict the near-fields in periodic, high dielectric contrast nanostructure arrays, and that it can combine with gradient-based algorithms to dramatically accelerate the local and global freeform optimization of diffractive photonic devices by orders of magnitude faster speeds. We anticipate that physics-augmented deep neural networks will transform the practice of nanophotonics simulation and design.